CN117594117A - Drug-target interaction prediction method based on heterogeneous graph contrast learning - Google Patents
Drug-target interaction prediction method based on heterogeneous graph contrast learning Download PDFInfo
- Publication number
- CN117594117A CN117594117A CN202311646153.9A CN202311646153A CN117594117A CN 117594117 A CN117594117 A CN 117594117A CN 202311646153 A CN202311646153 A CN 202311646153A CN 117594117 A CN117594117 A CN 117594117A
- Authority
- CN
- China
- Prior art keywords
- node
- drug
- meta
- target
- path
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
- G16B15/00—ICT specially adapted for analysing two-dimensional or three-dimensional molecular structures, e.g. structural or functional relations or structure alignment
- G16B15/30—Drug targeting using structural data; Docking or binding prediction
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
- G16B40/00—ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Medical Informatics (AREA)
- Spectroscopy & Molecular Physics (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Chemical & Material Sciences (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biotechnology (AREA)
- Crystallography & Structural Chemistry (AREA)
- Pharmacology & Pharmacy (AREA)
- Public Health (AREA)
- Medicinal Chemistry (AREA)
- Epidemiology (AREA)
- Databases & Information Systems (AREA)
- Bioethics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method for predicting drug-target interaction based on heterogeneous graph contrast learning. And establishing a heterogeneous graph based on the interaction network and the similarity network, obtaining node characteristics through a meta-path view and a random walk view, and obtaining final characteristics through contrast learning. And finally, inputting the feature vectors of the medicine and the target point into a trained classifier to obtain a prediction score so as to judge the possibility of interaction between the medicine and the target point. The invention ensures that under the condition of limited real data, the complex interaction of abundant semantic information and heterogeneous data in the heterogram is fully utilized, provides accurate guidance for drug repositioning, and identifies potential interaction between the drug and a target point, thereby predicting side effects of the drug.
Description
Technical Field
The invention belongs to the technical field of data mining and heterogeneous graph networks, and particularly relates to a drug-target interaction prediction method based on heterogeneous graph comparison learning.
Background
Predicting interactions between a drug and its target is critical for drug discovery and repositioning, with the aim of defining the relationship between the drug and the target protein. In the early stages of drug development, it was necessary to identify and select appropriate protein targets that play a critical role in the development and progression of disease, such as receptors, enzymes, transcription factors, and the like. Drugs exert their effects by binding to specific in vivo proteins (i.e., targets). Correctly predicting interactions between drugs and targets is extremely important for understanding the mechanism of action of drugs, optimizing drug design, reducing risk of side effects, and improving therapeutic effects. Scientists have employed a variety of bioinformatic and computational chemistry methods, such as molecular docking simulation, ligand methods, and machine learning, to predict binding affinity and possible binding patterns between drug molecules and target proteins. The method of molecular docking simulation aims at simulating the binding process of a drug to a target protein by predicting the structure of a receptor-ligand complex, wherein the receptor and the ligand are the target protein and a small drug molecule, respectively. However, the performance of these methods in predicting drug-target interactions depends on the availability of target structure information. To address this problem, a number of machine learning-based approaches have been proposed that exploit the bioinformatic similarity of drugs and targets to identify unknown drug-target interactions. Machine learning-based methods predict the presence of drug-target interactions by converting such interactions into classification tasks, and have been widely used for drug-target interaction prediction. The key assumption for these approaches is that similar drugs may have similar targets and vice versa. However, these machine learning based methods often do not provide satisfactory predictive performance, primarily because they rely excessively on the characteristics of given data, which may confuse the classifier, making it unable to make accurate predictions.
In recent years, the method based on the graph neural network is widely applied in the field of bioinformatics because the abundant topological information in the biological network and the complex interaction between the medicine and the protein can be considered at the same time. These methods learn representative features more efficiently, but face challenges in that the data for known drug-target interactions are limited and are costly to obtain. Recently, some biological studies have applied graph contrast learning to drug-drug interactions and drug-target interaction prediction tasks. These studies do not fully exploit the rich semantic information in iso-composition and the complex heterogeneous data interactions. Therefore, biological knowledge, structural information and interaction of the drug and the target are integrated into one network, so that the association relationship between the drug and the target can be more comprehensively captured, more information is provided for drug-target interaction prediction, and the prediction performance is improved.
Disclosure of Invention
The present invention aims to overcome the problems of very limited drug-target interactions and high acquisition costs in the real world. Limited data causes the problem that the generalization capability, the robustness and the accuracy of a drug-target interaction prediction model trained under supervised learning are not high. Therefore, we propose a method for predicting drug-target interaction based on hetero-graph contrast learning, which can provide reliable drug-target interaction prediction results and provide potential information for drug discovery and drug repositioning.
In order to achieve the above object, the method for predicting drug-target interaction based on hetero-graph contrast learning is characterized in that three interaction networks (drug-drug interaction network, drug-target interaction network, target-target interaction network) and two similarity networks (drug-drug similarity network, target-target similarity network) are firstly constructed, and a hetero-network can be constructed by combining these information together. In the meta-path view, the feature representation of the drug, target node, is extracted from the meta-path based neighbors by applying the graph rolling network and semantic level attention mechanism. In the random walk view, a sequence of nodes of drugs and targets is generated by performing random walks in the network. The contrast learning module then learns the representations of the nodes from the two views for each node, optimizing the node representations by computing similarities between the node representations. After the node representation is obtained, a gradient-lifting decision tree classifier is used to predict drug-target interactions. The method specifically comprises the following steps:
(1) And (5) collecting data. And collecting data of the drug and the target, and simultaneously collecting data of drug-drug interaction, drug-target interaction, target-target interaction, drug-drug similarity and target-target similarity.
(2) And preprocessing data. The characteristic representation of the drug node is obtained by extracting the molecular fingerprint of the drug and dimension reduction treatment. The characteristic representation of the target is obtained by extracting the relative proportion of amino acid and dipeptide in the protein sequence and performing dimension reduction treatment. The weights of the similarity network edges are determined based on the similarity scores. Constructing a drug-target interaction network G.
(3) Constructing a drug-target prediction model. The specific implementation of the step (3) is drug-target interaction prediction based on heterogeneous graph contrast learning, and the prediction is carried out through the following specific steps.
(3.1) constructing a meta-path view. And through the meta-path view, weighting and aggregating through the characteristics of the node neighbors in each meta-path, and then obtaining the final representation of the node in the meta-path view through aggregating the characteristics of a plurality of meta-paths. Wherein the aggregate formula in each meta-path is:
wherein P is n Represents the type of meta-path, h j And h i Representative node v i And v j Is characterized in that,is node v i In meta-path P n Is the current neighbor, lGCN layer number, < >>Is node v i And v j Respectively in meta-path P n Inverse of the product of the number of neighbors.
(3.2) constructing a random walk view. By means of a random walk view, a large number of node sequences can be generated starting from one starting node. The node sequence is then converted to a low-dimensional vector representation using skip-gram techniques. Wherein given a random walk sequence, the next node is predicted to be v i The probability formula of (2) is:
Pr(u i |φ(u 0 ),…,φ(u i-1 ))
wherein a random walk sequence v 0 To v i-1 Represented as { v } 0 ,…,v i-1 }。
(3.3), multi-view contrast learning. The low-dimensional vectors obtained from the meta-path view and the random walk view are compared to optimize the node representation. The loss function of the contrast learning module is as follows:
where a is the type of node,is a set of type a nodes, sim (u, v) represents cosine similarity between two vectors u and v, τ is a temperature coefficient, +.>Representing all samples, pos represents a positive sample set. />And->At the random walk view and the meta-path view, respectivelyThe node characterization obtained above.
(3.4) constructing a gradient lifting decision tree classifier.
(4) And sending the optimized node representation into a gradient lifting decision tree, and training a gradient lifting decision tree classifier based on the feature representations.
(5) And (5) model testing. And predicting the unknown drug-target point pair by using the trained classifier. And evaluating the prediction result according to the real data.
Drawings
FIG. 1 is a flow chart of a method for predicting drug-target interaction based on hetero-graph contrast learning;
FIG. 2 is a schematic diagram of a model structure of a drug-target interaction prediction method based on hetero-graph contrast learning according to the present invention;
Detailed Description
The following description of the embodiments of the invention is presented in conjunction with the accompanying drawings to provide a better understanding of the invention to those skilled in the art. It is to be expressly noted that in the description below, detailed descriptions of known functions and designs are omitted here as perhaps obscuring the present invention.
FIG. 1 is a flow chart of a method for predicting drug-target interaction based on hetero-graph contrast learning.
In this embodiment, as shown in fig. 1, the method for predicting drug-target interaction based on hetero-graph contrast learning of the present invention comprises the following steps:
s1: data collection drug-target interaction data from the drug bank database consisted of 969 drugs, 613 proteins and 11107 validated drug-protein interactions. Drug-drug interaction data was extracted from the drug bank database. Target-target interaction data is extracted from the HPRD database. The drug node data is extracted from the drug bank database. The target node data is extracted from the HPRD database.
S2: data preprocessing
The method comprises the steps of constructing a correlation network of the drug and the target respectively, initializing node characteristics and constructing a heterogeneous network, and providing a data base for drug target interaction prediction.
First, the drug-drug similarity network is calculated based on the chemical structure of the drug. Specifically, by calculating the chemical structural similarity between drugs, a drug-drug similarity score can be obtained. The calculated similarity score may be used as a weight for edges in a drug-drug similarity network.
The target-target similarity data is calculated based on the genomic sequence of the target. By comparing genomic sequence similarity between targets, a target-target similarity score can be obtained. Common calculation methods include Smith-Waterman algorithm, etc. The calculated similarity score may be used as a weight for edges in the target-target similarity network.
And secondly initializing the characteristics of the nodes. Initializing the characteristics of the drug node: first, a molecular fingerprint of a drug is obtained by extracting a SMILES expression of the drug. The molecular fingerprint of the drug is then subjected to a dimension reduction process, which is reduced to 128 principal components using principal component analysis, to reduce unimportant molecular descriptors.
Node feature initialization with respect to protein. By extracting the relative proportions of amino acids and dipeptides in the protein sequence, 420 descriptors were constructed. Then, the descriptors of the protein nodes are subjected to dimension reduction processing, and the descriptors are reduced to 128 principal components by using principal component analysis.
Through the edges and the node information, a heterogeneous network consisting of the nodes and the edges can be constructed for subsequent drug-target point interrelation prediction tasks.
S3: constructing a drug-target predictive model
The model is composed of four parts, namely a meta-path view, a random walk view, a contrast learning and gradient lifting decision tree classifier.
S3.1, meta-path view. Semantic information between different nodes in a heterogeneous network is captured by defining different meta paths. For each meta-path, a meta-path based graph rolling network is applied to extract feature representations from the neighbors of the meta-path. Specifically, for each node, by aggregating the feature representations of neighboring nodes connected to the node through a meta-path, the feature representation of the node in the meta-path view may be derived. In the aggregation process, a semantic-level attention mechanism is applied, and the characteristics of the neighbor nodes are weighted and aggregated according to the importance of different element paths. Acquisition of node representation: by aggregating the feature representations of the multiple meta-paths, a final node representation can be obtained. Wherein the aggregate formula in each meta-path is:
wherein P is n Represents the type of meta-path, h j And h i Representative node v i And v j Is characterized in that,is node v i In meta-path P n Is the current GCN layer,/-the neighbors of (1)>Is node v i And v j Respectively in meta-path P n Is the inverse of the number of neighbors.
S3.2, random walk view, by starting from a starting node, walk according to the random selection of neighbor nodes, repeat the process until the preset step number is reached. By performing the random walk multiple times, a large number of node sequences can be generated. The generated node sequence is treated as a sentence and then converted into a low-dimensional vector representation using skip-gram techniques. Wherein given a random walk sequence, the next node is predicted to be v i The probability formula of (2) is:
P r(v i |v 0 ,…,v i-1 )
wherein a random walk sequence v 0 To v i-1 Represented as { v } 0 ,…,v i-1 }。
And S3.3, comparing the learning modules, and maximizing the similarity of the node representations in the positive samples to optimize the node representations. Positive samples are highly correlated node pairs connected by multiple meta-paths, and negative samples are other node pairs. By maximizing the similarity of two node representations in a positive sample, the node representation can be optimized. These contrast learning optimized node representations can be used for subsequent drug-target interaction prediction tasks. The loss function of the contrast learning module is as follows:
where a is the type of node,is a set of type a nodes, sim (u, v) represents cosine similarity between two vectors u and v, τ is a temperature coefficient, +.>Representing all samples, pos represents a positive sample set. />And->The node characterizations obtained on the random walk view and the meta-path view, respectively.
S3.4: gradient lifting decision tree classifier. Firstly, taking node representations of the obtained medicine and target point as input, and constructing a training data set E train And test data set E test . Training data set E train For training a gradient-lifting decision tree classifier, test dataset E test For predicting the interaction of an unknown drug-target pair. For each drug-target pair<v i ,v j >Drug v i And target v j Is spliced to form a feature orientationAnd an amount h.
S4: model training: using training dataset E train And a feature vector h as input, to improve predictive performance by iteratively training a plurality of decision trees. The classifier trains a new decision tree in each iteration based on the differences (gradients) between the previous predictions and the true labels. Finally, the predicted results of the plurality of decision trees are combined to obtain a final predicted result.
S5: model test: for test dataset E test Each drug-target pair in (a)<v i ，v j >And inputting the feature vector h into a trained classifier to obtain a prediction score r. The predictive score r indicates the likelihood of interaction between the drug and the target, and a larger r value indicates that the drug and the target are more likely to interact.
In the invention, aiming at the problems that data is limited in real situation, the acquisition cost is high, and the traditional training strategy does not fully utilize the abundant semantic information in the heterogram and the complex interaction of heterogeneous data, a strategy for predicting the interaction of medicine and target points based on heterogram comparison learning is provided. The invention innovates on key technologies such as heterogeneous graph neural network and medicine-target point interrelation prediction.
While the foregoing describes illustrative embodiments of the present invention to facilitate an understanding of the present invention by those skilled in the art, it should be understood that the present invention is not limited to the scope of the embodiments, but is to be construed as protected by the accompanying claims insofar as various changes are within the spirit and scope of the present invention as defined and defined by the appended claims.
Claims (2)
1. The medicine-target interaction prediction method based on heterogeneous graph contrast learning is characterized by comprising the following steps of:
(1) And (5) collecting data. And collecting data of the drug and the target, and simultaneously collecting data of drug-drug interaction, drug-target interaction, target-target interaction, drug-drug similarity and target-target similarity.
(2) And preprocessing data. The characteristic representation of the drug node is obtained by extracting molecular fingerprints of the drug and performing dimension reduction treatment, the characteristic representation of the target spot is obtained by extracting the relative proportion of amino acid and dipeptide in a protein sequence and performing dimension reduction treatment, and the weight of the similarity network side is determined according to the similarity score so as to construct a drug-target spot interaction network G.
(3) Constructing a drug-target prediction model. The specific implementation of the step (3) is drug-target interaction prediction based on heterogeneous graph contrast learning, and the prediction is carried out through the following specific steps.
(3.1) constructing a meta-path view. The method comprises the steps of weighting and aggregating through a meta-path view through the characteristics of node neighbors in each meta-path, and then obtaining final representation of nodes in the meta-path view through aggregating the characteristics of a plurality of meta-paths, wherein an aggregation formula in each meta-path is as follows:
wherein P is n Represents the type of meta-path, h j And h i Representative node v i And v j Is characterized in that,is node v i In meta-path P n Is the current GCN layer,/-the neighbors of (1)>Is node v i And v j Respectively in meta-path P n Inverse of the product of the number of neighbors.
(3.2) constructing a random walk view. From a starting node, a large number of node sequences can be generated by random walk views, and then converted into a low-dimensional orientation using skip-gram techniquesThe quantity is expressed, wherein given a random walk sequence, the next node is predicted to be v i The probability formula of (2) is:
Pr(u i |φ(u 0 ),…,φ(u i-1 ))
wherein a random walk sequence v 0 To v i-1 Represented as { v } 0 ,…,v i-1 }。
(3.3), multi-view contrast learning. And comparing the low-dimensional vectors obtained by the meta-path view and the random walk view to optimize node representation, wherein the loss function of the comparison learning module is as follows:
where a is the type of node,is a set of type a nodes, sim (u, v) represents cosine similarity between two vectors u and v, τ is a temperature coefficient, +.>Representing all samples, pos representing the positive sample set, +.>And->The node characterizations obtained on the random walk view and the meta-path view, respectively.
(3.4) constructing a gradient lifting decision tree classifier.
(4) And sending the optimized node representation into a gradient lifting decision tree, and training a gradient lifting decision tree classifier based on the feature representations.
(5) And (5) model testing. And predicting an unknown drug-target point pair by using a trained classifier, and evaluating a prediction result according to real data.
2. The method of claim 1, wherein in step (3), the modeling of the graph neural network:
the model is divided into four parts, namely a meta path view, a random walk view, contrast learning and gradient lifting decision tree classifier; a meta-path view captures semantic information between different nodes in a heterogeneous network by defining different meta-paths, for each meta-path, applying a meta-path based graph rolling network to extract feature representations from the neighbors of the meta-path. Specifically, for each node, by aggregating the feature representations of neighboring nodes connected to the node through a meta-path, the feature representation of the node in the meta-path view may be derived. In the aggregation process, a semantic-level attention mechanism is applied, and the characteristics of the neighbor nodes are weighted and aggregated according to the importance of different element paths. Acquisition of node representation: by aggregating the feature representations of the multiple meta-paths, a final node representation can be obtained. Wherein the aggregate formula in each meta-path is:
wherein P is n Represents the type of meta-path, h j And h i Representative node v i And v j Is characterized in that,is node v i In meta-path P n Is the current GCN layer,/-the neighbors of (1)>Is node v i And v j Respectively in meta-path P n Is the inverse of the number of neighbors.
A random walk view, by starting from a starting node,the walk is performed in a manner that randomly selects neighbor nodes, and the process is repeated until a predetermined number of steps is reached. By performing random walk multiple times, a large number of node sequences can be generated, which are treated as sentences, which are then converted into a low-dimensional vector representation using skip-gram techniques. Wherein given a random walk sequence, the next node is predicted to be v i The probability formula of (2) is:
Pr(v i |v 0 ，…,v i-1 )
wherein a random walk sequence v 0 To v i-1 Represented as { v } 0 ,…,v i-1 }。
The comparison learning module maximizes similarity of node representations in the positive samples to optimize the node representations. Positive samples are highly correlated node pairs connected by multiple meta-paths, and negative samples are other node pairs. By maximizing the similarity of two node representations in a positive sample, the node representation can be optimized. These contrast learning optimized node representations can be used for subsequent drug-target interaction prediction tasks. The loss function of the contrast learning module is as follows:
where a is the type of node,is a set of type a nodes, sim (u, v) represents cosine similarity between two vectors u and v, τ is a temperature coefficient, +.>Representing all samples, pos representing the positive sample set, +.>And->The node characterizations obtained on the random walk view and the meta-path view, respectively.
The gradient lifting decision tree classifier firstly takes node representations of the obtained medicine and target point as input to construct a training data set E train And test data set E test Training data set E train For training a gradient-lifting decision tree classifier, test dataset E test For predicting the interaction of unknown drug-target pairs, for each drug-target pair<v i ,v j >Drug v i And target v j Is spliced to form a feature vector h.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311646153.9A CN117594117A (en) | 2023-12-04 | 2023-12-04 | Drug-target interaction prediction method based on heterogeneous graph contrast learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311646153.9A CN117594117A (en) | 2023-12-04 | 2023-12-04 | Drug-target interaction prediction method based on heterogeneous graph contrast learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117594117A true CN117594117A (en) | 2024-02-23 |
Family
ID=89919887
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311646153.9A Pending CN117594117A (en) | 2023-12-04 | 2023-12-04 | Drug-target interaction prediction method based on heterogeneous graph contrast learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117594117A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118197402A (en) * | 2024-04-02 | 2024-06-14 | 宁夏大学 | A method, device and apparatus for predicting drug-target relationship |
| CN118430640A (en) * | 2024-04-29 | 2024-08-02 | 湖南科技大学 | Medicine and target interaction prediction method based on multi-domain isomerism map polymerization learning |
| CN119580827A (en) * | 2025-02-08 | 2025-03-07 | 浙江大学 | Drug-target binding prediction method based on variational coding |
-
2023
- 2023-12-04 CN CN202311646153.9A patent/CN117594117A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118197402A (en) * | 2024-04-02 | 2024-06-14 | 宁夏大学 | A method, device and apparatus for predicting drug-target relationship |
| CN118430640A (en) * | 2024-04-29 | 2024-08-02 | 湖南科技大学 | Medicine and target interaction prediction method based on multi-domain isomerism map polymerization learning |
| CN119580827A (en) * | 2025-02-08 | 2025-03-07 | 浙江大学 | Drug-target binding prediction method based on variational coding |
| CN119580827B (en) * | 2025-02-08 | 2025-06-06 | 浙江大学 | Drug target binding prediction method based on variant coding |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107862173B (en) | A method and device for virtual screening of lead compounds | |
| JP7490168B1 (en) | Method, device, equipment, and medium for mining biosynthetic pathways of marine nutrients | |
| CN117594117A (en) | Drug-target interaction prediction method based on heterogeneous graph contrast learning | |
| CN113936735A (en) | A method for predicting the binding affinity of drug molecules to target proteins | |
| Liao et al. | Class-wise graph embedding-based active learning for hyperspectral image classification | |
| Li et al. | AdaDC: Adaptive deep clustering for unsupervised domain adaptation in person re-identification | |
| CN106021990B (en) | A method of biological gene is subjected to classification and Urine scent with specific character | |
| Wekesa et al. | A deep learning model for plant lncRNA-protein interaction prediction with graph attention | |
| Xu et al. | Multistructure graph classification method with attention-based pooling | |
| CN108763865A (en) | A kind of integrated learning approach of prediction DNA protein binding sites | |
| Demirel et al. | Meta-tuning loss functions and data augmentation for few-shot object detection | |
| CN117976035A (en) | A protein SNO site prediction method based on deep learning network with fusion features | |
| CN119724331A (en) | A method for predicting drug-target interactions based on geometric representation | |
| Liu et al. | Effective drug-target affinity prediction via generative active learning | |
| Fan et al. | Multi-view local hyperplane nearest neighbor model based on independence criterion for identifying vesicular transport proteins | |
| Zhang et al. | FMCA-DTI: a fragment-oriented method based on a multihead cross attention mechanism to improve drug–target interaction prediction | |
| Tan et al. | Protein representation learning with sequence information embedding: Does it always lead to a better performance? | |
| Li et al. | M4: Multi-proxy multi-gate mixture of experts network for multiple instance learning in histopathology image analysis | |
| CN119694442A (en) | A drug interaction prediction method based on dual-view graph neural network | |
| Wang et al. | Kinship similarity for open sets | |
| Bai et al. | A unified deep learning model for protein structure prediction | |
| CN117131214A (en) | Zero sample sketch retrieval method and system based on feature distribution alignment and clustering | |
| Thareja et al. | A novel optimized deep learning method for protein-protein prediction in bioinformatics | |
| Taju et al. | Using deep learning with position specific scoring matrices to identify efflux proteins in membrane and transport proteins | |
| Guo et al. | 3D genome assisted protein–protein interaction prediction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |